Research

2025 Catalysts for Progress? Mapping Policy Insights from Energy Research. Energy Research & Social Science. (with Brian Boyle, Stefan Müller, Sarah King and Robin Rauner)

2024 Electoral Reform and Fragmented Polarization: New Evidence from Taiwan Legislative Roll Call. Legislative Studies Quarterly.

2024 (Mis)perception of Party-voter Congruence and Satisfaction with Democracy. Political Science Research and Methods. (with Royce Carroll and Li Tang)

2023 The Role of Rituals in Adversarial Parliaments: An Analysis of Expressions of Collegiality in the British House of Commons. (Invited Contribution) Historical Social Research. 48 (3): 209-234. (with David Beck and Thomas Saalfeld)

2015 The Rationale for Supporting Nuclear Power: Analysis of Taiwanese Public Opinion Survey. International Relations of the Asia-Pacific. 15 (1): 147-176. (with Xiaochen Su, Chung-li Wu, Tai-De Lee, and Chen Tsao)

Peer-Reviewed Articles (in Chinese)

2025 官僚「再詮釋」領導人意識形態初探：以《人民日報》習近平外交思想的評論為例 (Bureaucratic ‘Reinterpretation’ of Leaders’ Ideologies: A Case Study of People’s Daily’s Commentary on Xi Jinping Thought on Diplomacy). 中國大陸研究 Mainland China Studies. (with Yi-Nung Tsai)

• Electoral Systems and Geographically-Targeted Oversight: Evidence from Taiwan Legislative Yuan. (with Li Tang) Minor correction, Electoral Studies

  Electoral systems fundamentally shape how legislators communicate with constituents and exercise representative functions. This paper examines how Taiwan’s electoral system reform — transitioning from multi-member districts under the Single Non-Transferable Vote (SNTV) to Single-Member Districts under a Mixed Member Majoritarian (MMM) — affects district legislators’ particularistic behaviors. Using fine-tuned transformer architectures, we analyze over 63,748 parliamentary questions from 402 district legislators across two decades to identify geographically targeted content. Controlling for legislator-year and socioeconomic attributes, we find that the transition from SNTV to MMM generally decreases legislators’ submission of geographically targeted questions, though this effect varies substantially across municipalities with different degrees of socioeconomic characteristics. Our findings suggest that SNTV demonstrates greater particularistic responsiveness to socioeconomic characteristics and variations than the MMM, demonstrating how different candidate-centered electoral systems produce different behavioral incentives.

• Estimating Factions of Red Guard under Mao’s China: A Slogan-based Text Scaling Method with Historical Documents. (with Yi-Nung Tsai) Invited to Revise and Resubmit
  Research on Red Guard publications during China's Cultural Revolution offers crucial insights into the period's political and cultural dynamics. We introduce a novel text analysis approach that addresses the challenges of analyzing non-spaced languages in historical contexts, advancing beyond traditional unsupervised text scaling –Wordfish– applications. Utilizing the Chinese Cultural Revolution Database, we combine keyword extraction techniques with Wordfish to estimate Red Guard units' ideological positioning. While our proposed approach largely aligns with historical accounts, we also reveal inconsistencies, particularly in how some Rebel-leaning Red Guard units rhetorically and ideologically act as fellow travelers with the Conservative coalition, deviating from established narratives and expert assessments. Our paper not only provides new insights into factional dynamics during the Cultural Revolution but also offers social scientists a new approach to studying Chinese politics and historical archives.

• Electoral Reform and Issue Attention in Legislative Oversight: From SNTV to Mixed-Member Majoritarian System in Taiwan. (with Yu-Ceng Liao and Yi-ting Wang)
  

  This research note examines how electoral reform — shifting from Single Non-Transferable Vote (SNTV) in multi-member districts to a mixed-member majoritarian system (MMM) — affects how closely legislators align with their party's policy attention. Using interpellations from 1999-2019 annotated with 422 fixed topic keywords, we estimate legislators' issue attention behavior across time using the Wordfish scaling method. We find that legislators under MMM prioritize more similar issues to their co-partisan members, while those under SNTV show greater divergence. Moreover, under SNTV, legislators from economically disadvantaged municipalities are more susceptible to local economic and demographic conditions. These findings indicate that SNTV creates stronger incentives for legislators to respond to local conditions, while MMM homogenizes legislative behavior across regional economic or demographic differences.

• Selecting and Validating Classifiers for Multilingual Stance Detection. (with Stefan Müller)
  

  Measuring stances on specific policies provides valuable insights for understanding policy-making, changes in political preferences, and party competition. In this paper, we fine-tune three multilingual transformer machine learning models based on annotated texts of stances in over 53,000 comments on Twitter and more than 67,000 comments to 150 political questions in German, French, and Italian. We compare Sentence-BERT, Multilingual BERT, and XLM-RoBERTa and show that these transformers can classify stances in several languages. After identifying the most suitable fine-tuned model, we compare the automated stance classification with hand-annotated evaluations of politicians' support for the annual budget, social media posts about protest events, and stances across policy areas. Drawing from our systematic comparison and validation across three cross-lingual transformer architectures, we provide recommendations for researchers when applying stance detection models to political texts.

• Multi-Agent Systems with LLMs for Synthetic Survey Experiments on Misinformation: Design, Implementation, and Limitations (with Linette Lim and Slava Jankin)

  In the digital age, misinformation research faces methodological and ethical challenges, especially when exposing human subjects to harmful content. To address these concerns, we create a Taiwan-based case, where elections are often influenced by misinformation under a complex geopolitical context. We build a multi-agent system (MAS) with LLM-operated agents based on anonymized Taiwanese survey data, implemented via AG2 (formerly AutoGen)—a framework for constructing AI agents and applications. Our simulated pseudo-lab environment enables the study of misinformation susceptibility safely through LLM agent interactions, removing direct ethical risks to human participants. We find that agents with stronger pro-China attitudes exhibit greater vulnerability to misinformation, particularly narratives fostering skepticism toward the United States. We further document our implementation and the system’s construction process—from data preprocessing and agent initialization to behavioral calibration and expert validation. Key implementation details, including environment setup, persona design, computational trade-offs, and scalability considerations, are provided. Finally, we make our prototype and AG2 implementation publicly available.

• Political Text Analysis with Embedding Regression: From Multilingual to Cross-lingual Application. (with Chen Zheng, Winnie Xia and Slava Jankin)

  This research note builds upon existing embedding regression techniques (i.e., Rodriguez, Spirling and Stewart, 2023a,b; Wirsching et al., 2025) to systematically compare different embedding architectures for political text analysis. We examine three types: static (fastText and BPE), sequential contextual (LSTM-based architectures: Forward, Backward, and Forward+Backward), and dynamic embeddings (Transformer-based architectures: XLM-RoBERTa and mBERT). We analyze differences between these three types using Benoit et al. (2016)’s coal debates from Members of the European Parliament, available in English, German, Spanish, Italian, Polish, and Greek. Our experiments demonstrate that XLM-RoBERTa, Backward, Forward+Backward, and BPE achieve better performance in predicting political stance on coal policy, with stable cross-lingual flexibility and consistency suitable for comparative political analysis across multilingual settings. While XLM-RoBERTa and bidirectional sequence models maintain the highest accuracy, BPE offers an optimal balance of performance and computational efficiency. We are currently packaging this workflow as open-source software.

• Game-Theoretic Multi-Agent Systems with LLMs for Crisis Negotiation and Simulation. (with Shuli Zhang)
  

• A Micro-Founded Model of Canvassing, Ideological Distance, and Political Misperception (with Li Tang)

• Multi-Dimensional Policy Congruence and Political Attitudes (with Royce Carroll and Li Tang)

• Vicar of Bray: Performative Loyalty and Career Survival in Maoist China. (with Yi-Nung Tsai)

• From Legislators to Mayors: Political Career and Distributive Politics in Taiwan Municipalities.